Random Projection for Indirect Quantization

Abstract

In indirect quantization tasks, the quantize-thenestimate (QE) scheme is commonly adopted when encoders are distributed or task-agnostic. In some scenarios, however, the observation dimension substantially exceeds that of the underlying state, rendering direct quantization of the observations highly inefficient. In this work, we propose a random-projectionthen-quantization (RPQ) scheme. Specifically, high-dimensional observations are first projected onto a lower-dimensional representation via random projection, after which quantization is applied only to the reduced dimensions rather than to all original dimensions. This strategy reduces the required bit rate and thereby alleviates communication overhead. The rate-distortion performance of the proposed scheme is analyzed. Under low-rate constraints, the proposed scheme outperforms the conventional QE scheme and, in certain scenarios, continues to do so even at higher rates.